Summary

The dsPIC30F Speech Recognition Library provides voice control of embedded applications that require an alternative user interface. With a vocabulary of up to 100 words, the Speech Recognition Library allows users to control their application vocally. The Speech Recognition Library is an ideal front end for hands-free products such as modern appliances, security panels and cellphones. The Speech Recognition Library has very modest memory and processing requirements and is targeted for the dsPIC30F5011, dsPIC30F5013, dsPIC30F6012 and dsPIC30F6014 processors.

Description

The Speech Recognition Library provides isolated, speaker independent word recognition of US English. It allows a user to control an application through a set of fixed, voice commands. The library has already been pre-trained by a demographic cross-section of male and female US English speakers. Conveniently, no training is required for end-users of the product.

The library samples speech data from a voice codec connected to the dsPIC30F's Data Converter Interface. The data is processed a frame at a time, and when a word ending is detected, the received word is identified using Hidden Markov Model processing. After the library identifies the word, your application may then take some pre-defined action.

The Speech Recognition algorithm is written in assembly language to optimize performance and minimize RAM usage. A well defined API makes it easy to integrate the Speech Recognition Library with your application. Library functions let your application easily disable and enable speech recognition. The library lets your other system processing operations take place without disrupting speech recognition.

A PC-based Word Library Builder program allows you to build a custom library from a Master Library of 100 common words. A noise profile is selected that suits your operating environment. The noise profile consists of a noise type and a signal-to-noise ratio (SNR). The noise type can be any combination of 3 different noise sources (Babble, Volvo and White noise), and the SNR may be as low as 15 dB. The Word Library Builder program generates a source file that you then use to build your application. This file contains data tables that the library uses to perform speech recognition.

Features

Key features of the dsPIC Speech Recognition Library:

Devices Supported

Resource Requirements

Part Numbers and Ordering Information

Note: Quantities are per project, payable as a one-time license fee based on estimated lifetime volume for products resulting from the project. Please consult the factory for quantities above 100K.